At this time, two methods exist to calculate a pregnant woman's presumed delivery date (DPA) : one adds 280 days to last menstruation date (Naegele rule), other estimates early pregnancy's date by imagery and adds 270 days. Unless pathology requires a trigger, this DPA estimated a early pregnancy is not re-estimated. These methods are simple and arbitrary : Mongelli and al. in 1996 found that out of nearly 40 000 unique pregnancies, only 4% give birth at determined DPA by echography and 70% at more or less 5 days. Jukic and al. in 2013 they estimate a natural variation of 37 days between pregnancy durations. Face of these poor performances, the calculating DPA method seems to be open to improvement.

Thus, the DPA calculation formula does not take into account the individual patients characteristics (age, occupation, antecedents ...), nor the follow-up data collected during pregnancy. Jukic and al. in 2013 propose a first model with some individual characteristics and medical measures (period between ovulation and early pregnancy, hormone peak) to refine the estimation. Their study gives promising results but their small patients number (a hundred) does not allow them to detect all interactions. Moreover, their method calculation is not dynamic, i.e it does not refine the DPA as pregnancy progresses. To our knowledge, no studies developing an evolutionary model over time for the DPA exist. However, objectives of a more accurate estimate of expected date are multiple and important. The investigators will mention here the two main ones :

• A better understanding of mecanisms leading to early labour or abnormally long gestation in order to anticipate patients at risk
• A better material and human needs anticipation, allowing a more efficient organization more adapted to activity and a care of each parturient in optimal conditions.

Our study will focus on predictive model elaboration of pregnancy duration that will evolve as the pregnancy progresses and new data collected. The investigators are considering a machine learning methodology by patient's medical record computerization at the Groupe Hospitalier Paris Saint-Joseph (GHPSJ) since early 2016. Thus, for patients who gave birth from end of 2016, the investigators have a large amount of information on their pregnancy and follow-up on hospital servers, which motivates an automatic approach based on massive data analysis.

This study thus intends to implement advanced techniques in Machine Learning (Online Learning, Support Vector Machine ...) to advance a powerful calculation model.

Inclusion Criteria:

• Patient whose age ≥ 18 years old
• Patient who gave birth at GHPSJ maternity between 01/01/2017 and 02/28/2018

Exclusion Criteria:

• Patient who expressed her opposition to participate in the study
• Patient under guardianship or curatorship (unless consent is provided)
• Patient who gave birth at less than 32 weeks amenorrhea
• Pregnancy marked by MFIU (fetal death in utero)

Patient who gave birth at GHPSJ maternity between 01/01/2017 and 02/28/2018.